Chaotic Quantum Beetle Swarm Optimization Based Empowered Model of Bidirectional Long Short-Term Memory for Prediction of Survival of Paediatric Undergoing Hematopoietic Stem Cell Transplantation

Abstract

For certain patients with hematologic diseases, hematopoietic stem-cell transplantation (HSCT) is a therapeutically restorative surgery. Although the danger of transplantation has decreased recently, morbidity and death are still high, thus choosing who, what, and when to do a transplant is crucial. Machine learning models play a crucial role in forecasting the mortality rate of paediatric patients following a transplant, the problem of voluminous dataset and class imbalance affect the detection accuracy of the prediction model greatly. Hence, in this paper we developed a two stage algorithms for potential feature selection and optimizing the functionality of the deep learning classifier. To reduce the irrelevancy and improve the accuracy rate of the survival prediction of HSCT children, a novel stochastic differential evaluation method is developed in this work to determine the most informative features that contribute more information about the survivability of the paediatric patients. The problem of overfitting due to the class imbalance and the inappropriate assignment of hyperparameters in long short-term memory is handled by devising an empowered model of Bidirectional Long Short-Term Memory (BLSTM) which conducts the learning process twice, to understand the input patterns in depth. Still, the problem of weight decaying is a major problem in BLSTM due to lack of knowledge in assigning values to the hyperparameters this proposed work utilize the Beetle Swarm Optimization food searching behaviour by achieving global optimization search space with chaos mapping improves the performance of the BLSTM effectively. The experimental results provide the evidential proof of the proposed chaotic Beetle Swarm Algorithm empowered Bidirectional Long Short Term Memory (CBS-BLSTM) produced highest accuracy rate of 98% compared with other deep learning models.